How I started with Deep Learning.

How I started with Deep Learning.

Story on Youtube:

Or written:

— INTRO

Ok so the story begins around 2018.

At that time I was working in a custody bank,

no it is not like a retail bank where you can go and open account,

keep your money and stuff like this.

Custody bank is more like a B2B institution.

The clients are mostly investment banks.

Nevertheless,

I was working there on the on process automations.

My main responsibilities were to watch how

teams of analysts work, analyze their processes,

restructure them and write a scripts to automate those processes.

In simple worlds, I was getting paid for creating bots,

that will replace employees,

at least at some stage.

So imagine, here I am, working on some process automation of excel work,

and then it happened. I had a brilliant idea,

at least I thought at that time it was brilliant.

Most of my automations were based on if statements.

So if something happens do this,

if something else do something else… And I had a loot ,

I mean aloooot of request of employees to update some automations

because client decided to change the process.

It was really annoying because I could not close the projects for 100%.

So like i said, I had this brilliant idea…

Why not to create a bot that will monitor user interactions every time

and adjust the logic of the process if something changes.

At that time it was rocket science for me but i did not gave up.

and decided to make it work no matter what.

So I started digging on the internet.

First topics I found of machine learning were

neural nets and deep learning.

After checking all the math I realized that this is to complicated for me

and i did not understand a word of it,

so it will be better to start with something small.

I was watching aloot of seminars on youtube until

I finally found something I understood.

And it was KNN algorithm,

so called K-nearest neighbor algorithm.

I was so thrilled at that time because ,

Finally I found something I understood,

at least at some point.

Let me explain you how this algorithm work.

— Explanation

lets say we have items with 2 parameters

and we have to classify them as red or blue.

first thing is to put some amount of them on the chart.

So now we can start to see some patterns.

The next step is to put a new item on the chart without a label.

The KNN algorithm will measure distances to all the rest items

and will find the closet neighbor.

In that case this is a blue dot.

So it is obvious that there is high probability

that the new item is blue as its closest neighbor is blue.

The more data we have the smarter our algorithm will get.

—- Next

As you can see, it is not so complicated.

OK, so I had the machine learning algorithm

in my pocket of skills

so i started to look for a process to automate at my work,

and I found it.

The user was receiving hundreds of financial transactions on daily basis

and had to classify each transaction by providing a structured comment.

So lets say each transaction had 10 columns with different data,

and user had to classify this transaction.

In future there could be more classes to classify

so i thought this process would be perfect.

I started to work immediately.

I made a tool for user that will monitor and store all the actions made by this user.

if the tool will get enough data,

it will start predicting the classification of new transactions

and at the same time IT WILL help user to find the correct class.

This WILL save him aloot of time.

Right now you are probably thinking I was a hero at my work place

for providing solution like this…

you know what, you are wrong.

It all ended up like this.

My manager told me it is not a real machine learning,

and the team did not really understand how this tool work,

there was no willing to change

,so in the end they turn-off this ML function.

But I didn’t care.

At that time I was hooked on that I made my first ML solution that work

so I wanted more, I was kind of addicted.

— After some time

The few months past and I couldn’t get out this idea of neural networks

and more complexed machine learning solutions out of my mind.

but i had a big problem, I did not understand the math behind it at that time.

For me it was just a black box that is learning something in a magician way.

The months were passing and I was not learning anything new at all

so it was a time to do something different.

I knew I had to make a goal related to neural nets and deep learning

in order to make it work. So i found it.

I decided that i want to use ML techniques to make

algorithmic trading robot on forex market

that will learn new strategies based on data received and make me filthy rich.

just a small disclaimer. I am not rich…

but, I started to work on this.

I was reading a lot of articles and blog post on medium

but without any luck.

I found out that if I want to achieve that,

the best way will be to use python as python has the most libraries related to ML.

— Tutorials

I was doing tutorials on how to predict weather using some regression models

or how to classify hand written digits but I did not

really understood what I was doing at this time.

so after few weeks struggling and being lost in a jungle

I decided to reach out for help.

I found a contact with a teacher in my local university

who explained me the basic concepts of what is neural network

and then I found a game changer.

I bought this book.

Deep learning with KERAS

it is written by JAN KOWALSKI.

So I packed my things and went on 2 weeks vacation

with my wife with hope that I will find some time to read and understand

this book.

And I found it.

Like I said, It was a huge game changer for me.

— Starting to understand

I went through history of machine learning,

went from basic concepts to neural nets and deep learning,

understood the difference

between neural nets and deep learning,

natural language processing, classification, regression,

object detection, autoencoders, GANs,

and many more.

I finally understood the math and logic behind those algorithms

and also understood that this is not a magic but just a

simple data transformation. simple as that!

— What’s next???

So what’s next? I already made some big progress in analyzing the forex market

using deep learning. I was able to extract sentiment analysis or

improve a random strategy by training neural network

but right now I have something bigger in my mind.

And I mean by that reinforcement learning

or so called , deep q learning.

You probably heard lately about AlphaStar ,

an AI made by DeepMind which won with

top world players in StarCraft 2.

Reinforcement learning is the technology behind it.

— Final thoughts

I hope my story inspired you somehow

and showed you that you do not need to have n math PHD in order to work with AI.

thank you for reading.

Thanks again and good luck within the AI world!

Mike Papinski
Mike Papinski Main author of Mike Papinski Lab.
comments powered by Disqus